Achint Soni
Research, with a sense of possibility

Building models
that imagine.

I’m Achint Soni. I explore how generative models see, create, and understand the world — and how we can give them better control.

Explore my research View CV
Generative AI · Computer vision · Representation learning
A field of possibilities01 / ∞
Curiosity → possibility
Achint Soni Hi, I’m Achint.
A little more about me
Generative world modelsImage & video generation3D reconstructionDisentangled representations
01The person behind the papers

Driven by questions.
Drawn to building.

What does it take for a model to understand the world — and create something new within it?

My research spans generative world models, image and video generation and editing, 3D reconstruction, and disentangled representation learning. I’m especially interested in making generative models more controllable.

My graduate research in Computer Science at the University of Waterloo was advised by Prof. Sirisha Rambhatla and Prof. Charles Clarke.

Before that, I received my B.Tech in Electrical Engineering from IIT Kanpur, where I was advised by Prof. Faiz Hamid and Prof. Laxmidhar Behera.

Google Scholar GitHub
University of WaterlooComputer Science · Graduate researchResearch
IIT KanpurElectrical Engineering · B.TechFoundations
02Ideas, put to work

Selected research.

A few questions I’ve explored.
And the work they turned into.

6 works
EMNLP 2024 · Main05

VIDEOSCORE: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation

Xuan He, Dongfu Jiang, Ge Zhang, Max Ku, Achint Soni, et al. All authors +

Xuan He, Dongfu Jiang, Ge Zhang, Max Ku, Achint Soni, Sherman Siu, Haonan Chen, Abhranil Chandra, Ziyan Jiang, Aaran Arulraj, Kai Wang, Quy Duc Do, Yuansheng Ni, Bohan Lyu, Yaswanth Narsupalli, Rongqi Fan, Zhiheng Lyu, Yuchen Lin, Wenhu Chen

Better feedback. Better generation.Read paper
Find me on Google Scholar
03Along the way

Notes from
the journey.

Papers, milestones, and new beginnings.

  1. Research

    LOCATEdit accepted to ICCV 2025.

  2. Research

    VideoAgent accepted to the Workshop on Reinforcement Learning Beyond Rewards at RLC 2025.

  3. Research

    “Why Do Variational Autoencoders Really Promote Disentanglement?” accepted to ICML 2024.

  4. A new chapter

    Joined the David R. Cheriton School of Computer Science at the University of Waterloo as a graduate student.

  5. Milestone

    Graduated from IIT Kanpur with a B.Tech in Electrical Engineering, with minors in machine learning and theory of computation.

  6. A new beginning

    Received an offer of admission to the MMath in Computer Science program at the University of Waterloo.

04Good things start with a conversation

Something
on your mind?

A research question, a collaboration, or an interesting engineering challenge — I’d love to hear about it.

Let’s talk.
GitHub LinkedIn Scholar X / Twitter

LOCATEdit See the editing in action.

Graph Laplacian Optimized Cross Attention for Localized Text-Guided Image Editing